LAUSR

Abstract Drop-Foot is a common problem resulting from a range of neurological conditions, and prevents normal leg swing during gait, leading to abnormal, inefficient motion with an increased risk of falling. It damages the quality of life of over 122,000 people in the US and 11,400 people in the UK every year. Functional electrical stimulation (FES) addresses drop-foot by artificially contracting the tibialis anterior, and has had considerable success both clinically and commercially. However current commercial controllers are open loop and have long set-up times. The few controllers in the research domain are predominantly open-loop, lack accuracy, and struggle with muscle delays, non-linearities and the onset of fatigue. More advanced controllers require extensive sensor data and/or are highly dependent on an identified model. Recent developments have shown model based controllers combined with learning can facilitate higher accuracy, however previous attempts employed batch-wise learning, and led to disjointed control signals. This paper applies repetitive control (RC) to drop-foot for the first time, facilitating a continuous, smooth process of learning with no resetting. To maximise performance, a comprehensive extension to the traditional RC framework is undertaken to enable only isolated time points to be tracked, improving robustness and reducing memory and communication requirements. Experimental data confirms that RC can achieve normal gait when applied to FES-assisted gait with no voluntary effort. The new ‘point-to-point’ RC framework outperformed traditional RC, while using only 5 data points per gait cycle and minimal control effort.